From tracking recruitment to running it

AI is changing what hiring platforms can actually do. For years, recruitment software has talked about "AI" — but in practice that has usually meant fairly narrow tools: CV parsing, keyword matching, or occasionally automated calls. Much of it has been more hype than transformation.

What's emerging now is something quite different. Instead of simply analysing information, software is increasingly able to run parts of operational workflows itself. Not just reading CVs or summarising interviews, but executing tasks that previously required people: chasing references, collecting employment histories, following up on missing information, and keeping a clear record of everything that happens along the way.

That idea understandably makes people uncomfortable. The thought of software "running recruitment" can sound alarming, and human judgement remains essential in hiring decisions. But the reality is that most recruitment processes already involve large amounts of repetitive operational work. When technology can take care of those steps reliably, it frees people to focus on the parts of hiring where human judgement matters most.

The challenge, interestingly, isn't building the technology itself. It's shifting expectations. Most people still expect hiring platforms to track recruitment rather than execute it. As we've been building Lily — an AI-first hiring platform for frontline industries — that gap in expectations has turned out to be one of the biggest challenges.

Hiring decisions can be easy. Proving them rarely is.

In regulated industries like care, regulation means that when hiring someone you don't only need the best candidates — you also need to prove that you hired them fairly and safely. Regulators care that interviews were conducted consistently and recorded properly, that employment histories were checked, references were sought, gaps were investigated, and reasonable steps were taken to ensure someone was suitable for the role.

Most recruitment software treats these requirements as documents or checklists. Platforms give you places to upload references, store identity documents, and tick off compliance steps. But the responsibility for actually executing those steps still sits almost entirely with people. Managers are left to chase references. HR teams end up tracking down missing information, sometimes months after employment has already started.

If these busy individuals don't remember to log every activity manually, CQC inspections can quickly become stressful. Organisations often find themselves scrambling to reconstruct evidence from individual email accounts and document stores, trying to demonstrate that the right steps were followed.

Why ATS platforms became so configurable

Historically, this approach made sense. Software was good at organising information but not particularly good at performing operational work. Recruitment platforms were built to store data and track stages, while the operational work still sat with people.

To support different industries and hiring processes, these systems evolved into highly configurable platforms. Organisations could build their own workflows, forms, and compliance processes so the product could adapt to almost any environment.

The problem is that flexibility often turns into complexity for the people actually using the system.

That creates two problems. First, the systems themselves become far more complicated than they need to be. When everything is configurable there is simply more surface area in the product: more forms to create, more workflows to maintain, and more configuration to understand.

This becomes particularly obvious in regulated sectors. In care, for example, providers are ultimately trying to satisfy the same underlying guidance from the CQC around safe recruitment. Employment histories need to be checked, references need to be obtained, gaps need to be investigated, and interview decisions need to be documented.

Yet because hiring platforms rely on configurable workflows, every organisation ends up building its own version of how this should work. Slightly different forms. Slightly different workflows. Slightly different interpretations of what "compliant" looks like. The result is that dozens of organisations end up configuring slightly different versions of the same safe recruitment process — each trying to interpret the same regulatory guidance inside a system that was never designed to run the process itself.

Second, the expertise needed to support customers often sits in the wrong place. Customer success teams can explain how to build a form or configure a workflow, but they're rarely experts in sector-specific safe recruitment. The product knows how to configure things, but it doesn't necessarily know what good looks like. The result is complexity for customers and support teams trying to bridge a knowledge gap that arguably shouldn't exist.

What happens when the platform runs the workflow

References are a good example of how this plays out in practice.

Ask ten care providers how they handle references and you'll usually hear variations of the same conversation. Some ask for two references, others ask for three. Some prioritise the most recent employer, others try to reference every previous care role. Some send a single email and move on; others chase more persistently. Most systems simply provide a place to upload whatever references eventually arrive.

But if you step back and look at the outcome regulators care about, the shape of a sensible process becomes fairly clear. Two returned references including the most recent employer is widely considered the standard. Demonstrating that attempts were made to obtain references from earlier care roles is often just as important. What matters is not only the final documents but the evidence that reasonable effort was made to obtain them.

This is where AI begins to change what a hiring platform can actually do.

Instead of asking organisations to configure how references should work, Lily's platform runs the process itself. Lily requests references automatically, prioritising the most recent employer while still attempting to contact earlier care roles. If responses don't come back, the system follows up. Emails can be sent, reminders scheduled, texts triggered, and calls made where appropriate. Every attempt is logged. Every response is recorded.

Crucially, the system is not replacing human judgement. Hiring managers and recruiters still read the references, assess whether they're appropriate, and ultimately decide when the requirement has been satisfied. Their role becomes one of verification and approval rather than operational chasing. What changes is that they no longer need to worry about who has been contacted or how many times someone has been chased. They simply review what comes back and sign the step off as complete.

From documentation to evidence

What matters here is not just efficiency — although removing that administrative burden is valuable. What matters is that the system generates the evidence required to demonstrate safe recruitment. When an inspection happens, organisations no longer need to reconstruct the story of how references were obtained. Lily can show exactly what happened: who was contacted, how many times they were chased, which channels were used, which responses were received, and when the requirements were satisfied.

And once a system is doing this work consistently, something else becomes possible. You start to accumulate real operational data about how hiring actually works. How long references typically take to return. Which types of employers respond fastest. When follow-ups are most effective.

Having that data opens the door to improving the process over time. Lily's current capabilities already change how these workflows are executed — but the real opportunity with an agentic approach is that the process can continue to evolve and improve as we better understand how these operational steps actually behave in the real world.

When software runs workflows, it has to understand them

This shift has implications beyond technology. It changes what we should expect from hiring platforms, and it changes what product teams are responsible for building.

Building systems like this requires taking a much stronger stance. Configurable platforms push responsibility onto the customer. The product provides the tools and each organisation decides how to use them. Building a platform that actually runs recruitment means making clearer decisions about what good looks like.

That requires real domain understanding. Anyone building systems like this needs to understand the sector they're building for, the regulatory expectations that shape it, and how those expectations differ across geographies. Safe recruitment in UK care looks different to hiring in other industries or other countries.

Historically, that kind of specificity was something most software tried to avoid. Hard-coding regulatory logic into a product was seen as limiting. It made the system less flexible, reduced the potential market, and created ongoing maintenance work whenever rules changed. The safer approach was to provide configurable tools and let customers build their own processes.

But when software starts running operational workflows rather than simply storing information, that approach starts to break down. If the system is responsible for executing the process, it needs to understand what the process actually is. That means doing more of the hard work inside the product rather than pushing it onto the customer.

Historically that would have been a major constraint. Today, with modern development tooling and AI-assisted engineering, adapting and evolving software has become dramatically easier. Encoding domain logic is no longer the limiting factor it once was.

The future of hiring platforms is execution

For Lily, the goal is not to optimise for regulators. The goal is to make hiring faster, easier, and more reliable for employers in frontline industries. But when the hiring process itself is executed correctly and consistently, the compliance evidence naturally follows. Safe recruitment becomes a by-product of well-run hiring operations — rather than a separate administrative burden.

Recruitment software has spent the past twenty years tracking hiring. The next generation of hiring platforms will actually perform it.

Want to see how Lily runs the reference and compliance workflow in practice? Book a demo and we'll walk you through it.